When to use AI? The defining question for productivity and authenticity.

The argument against maximalism

Beyond gaming broken token incentives, using AI for everything has some major pitfalls.

The most obvious is the slop of delegating key human interaction to AI. Converting language through a chatbot, having agents conduct outreach without specifying what is AI and what is not. These actions can all ruin key trust and create borders that make alignment and connection more difficult.

But building with AI is great right?

I certainly think so. But there are quite a few traps that are under discussed here as well.

More is not always better. Building 100 prototypes to tackle a single problem does not guarantee a better outcome than 1. If you are solving a human problem, it's impossible to infinitely scale feedback. And sitting with one concept for a while to understand it deeply can be as valuable as spinning up infinite possibilities and using a non-perfect method to decide which one wins.

More without reward also leads to burnout. There are always more agents that can be running, apps that can be built, processes to be optimized. But while speed is necessary, turning into an AI coding vampire can damage mental health and degrade critical brain functions.

The argument for heavy adoption

At this point you are in the minority if you work in tech and do not leverage AI everyday.

The productivity boost is almost inarguable. And to the point of this article: how else are you going to learn when to use and not use AI without the experience of succeeding and failing with it?

The defining characteristic of our identities over the next few years could be what tasks, and with what level of guidance, we delegate to AI.

The cost structure

Everything comes at a cost. The 20 extra PRs written with AI give you an outsized productivity metric at the cost of less learning per commit and in most cases, some alignment drift.

On the other extreme, leveraging and learning AI to complete a new task has a cost as well. Most singular tasks can be achieved more easily using previously learned workflows rather than delegating to AI for the first time.

So what do you as a person and an employee prioritize? Productivity, productivity signaling, learning, alignment, or something else? It's complicated. But these are the same tradeoffs that always exist in enterprises and are amplified with new tech.

The feel slope

Frontier models are general. Used for so many different tasks that it is really hard to get a feel for what they can do.

Even within coding and design, me, my friends and coworkers all use different tools and structures to accomplish similarish tasks.

Many people start by reading posts about which models are good at what, which tactics work for types of building, then get frustrated when sources contradict.

The problem is that while prompting and guiding AI is simple in nature, the feel for what it can handle, when to intervene, when to front load on planning vs. let it run and correct later, is actually pretty hard. And regardless of which model or tool you use, it does not come right out of the box.

The skill curve

So we have established blindly using AI for everything is not great. But you need to use it a lot to get a feel for how to make it work for you. So how do you evolve and measure this skill?

Measuring is hard. There is no established meta for what augmented humans can do, and in many cases, the reward structure is lagging.

Evolving is easy. Make stuff. Try things. Follow building when it energizes you, put it away when other things are more urgent to your business or important to you.

I am fortunate to live in Austin, TX. A community with a ton of tech employees who are all learning these skills. My own Austin Build Club really helped me keep myself accountable and do more as well as get feedback from peers.

The authenticity of analog

Intentionally not using AI for things that it is good at can be a performance, a rebellion. It can also be the right decision.

Taking notes with a pen and paper. There is something you learn when your brain does this. For certain people in certain situations, it's bold, but it can be helpful.

Hand coding an app and learning JavaScript fundamentals rather than the agentic stack. For early developers, it's part of understanding the medium that is still key. For non-technical people, sure it takes a while, but it is another skill that compounds.

A personal choice

For me, communication is a hard no-AI-zone. I would rather be my authentic and less than perfect self than risk filtering out intention. For design engineering, everything generated always passes through multiple manual checks, tests, and edits.

But everyone is different, and the capabilities and structures around AI are always changing. Being intentional about what is better with AI, what is not, and why: that remains.